Counting tablets: manual, weight-based or camera?
Three methods dominate quantity checks in pharmaceutical production. They fail in different ways, and the failure modes matter more than the headline accuracy.
Published 2026-09-168 min read
In short
- Manual counting is flexible and slow, and its error rate rises exactly when the line is under pressure.
- Weight-based counting is fast and cheap, and it assumes a stable unit weight you have to prove.
- Camera counting sees every unit and leaves an image behind, which is what makes a result defensible later.
The job to be done
The task sounds trivial: confirm that a tray, kit or container holds the quantity it should. In regulated production it is not trivial, because the answer has to be documented, attributable and defensible months later during a review or an inspection.
That is why a comparison of counting methods cannot stop at accuracy. It has to include what each method leaves behind as evidence, and what it assumes about your product.
Manual counting
A person counts, a second person verifies, both sign. It needs no capital equipment and handles any product, any shape and any container on day one.
The cost shows up elsewhere. Counting several hundred small units is slow, and error rates rise with fatigue, interruption and time pressure. The documentation is a signature on a form: it records that a check happened, but not what was actually seen. If a deviation appears later, there is nothing to re-examine.
Weight-based counting
A balance infers the quantity from the total weight and a known average unit weight. It is fast, it is well understood, and for uniform units it is hard to beat on cost per check.
Its assumption is the unit weight, and that assumption has limits. Coating variation, moisture uptake, broken or partial units and mixed batches all move the total in ways a scale cannot separate from a count difference. The smaller the unit relative to the tare, the more the resolution of the balance decides the result. The evidence is a number and a printout; nobody can look back at the units themselves.
Camera counting
A camera images the units and a detection model counts them individually. It does not assume anything about weight, so broken or partial units can be a separate class rather than a rounding error, and the same capture can flag damage while it counts.
It has its own prerequisites. Units must be visible: spread in a single layer in a tray, fixture or cavity. Heavily piled or overlapping units hide each other from any camera, and no model recovers what the sensor never saw. Lighting has to be controlled, and a detection model has to be trained on the actual product, which is a task somebody has to own.
What it produces is different in kind: an annotated image per capture. Months later a reviewer can see what the station saw, which is the part that holds up in an investigation.
Side by side
| Manual | Weight-based | Camera | |
|---|---|---|---|
| Speed per check | Slow | Fast | Fast |
| Setup effort | None | Low | Model training and lighting |
| Depends on unit weight | No | Yes | No |
| Handles damaged units | If the operator notices | No | Yes, as a trained class |
| Needs units visible | No | No | Yes, single layer |
| Evidence left behind | A signature | A number | An annotated image |
| Scales to many stations | With headcount | Yes | Yes |
Choosing between them
- Choose manual
for low volumes, constantly changing products, or as the fallback while an automated method is being qualified.
- Choose weight-based
for uniform units with a stable, proven unit weight, where throughput matters more than per-unit evidence and the tare is small relative to the contents.
- Choose a camera
when the units are visible in a single layer, when you also want damage or completeness checks, or when your process needs an image to accompany the number.
The methods combine well. Several sites weigh in-line for throughput and use a camera at defined control points, where the record matters most.
The questions that decide it
How are the units presented at the point of the check: piled, in a tray, in cavities?
Is the unit weight stable enough to prove in a qualification, across batches and seasons?
Does a deviation need to be investigated later, and with what?
Who owns the method once it runs: production, QA or engineering?
What happens on a product changeover, and how long does it take?
Frequently asked
How accurate is camera counting in practice?
It depends on the detection model, the lighting and how the units are presented, not on a single datasheet number. Because the model is trained on your own product and every result is stored with its image, the honest way to establish accuracy is to measure it on your own trays during performance qualification.
Can a camera count units in a blister or a bottle?
Cavities such as blister pockets work well because every unit is separated and visible from above. Bottles do not: the units overlap and hide each other, which is a case for weighing or for counting before filling.
Is checkweighing not sufficient for release?
It often is, and many processes rely on it. The question is whether your risk assessment accepts an inferred count for the specific product, and whether the record it leaves is enough to investigate a deviation later.
What does a camera add beyond the count?
Two things: a per-unit inspection such as damage detection in the same capture, and an image that makes the result reviewable afterwards instead of a number with nothing behind it.
Sources
The primary texts behind this guide. Each link opens on the issuing body's own site, so you can check the wording yourself.